Translategemma-4B-It at Main (huggingface.co)

Google has announced the launch of TranslateGemma, a new family of lightweight, state-of-the-art translation models that can handle tasks across 55 languages. Built upon the Gemma 3 architecture, these models are designed for deployment in resource-constrained environments, making it easier for developers and researchers to access potent translation capabilities. TranslateGemma can translate both text strings and images, supporting a variety of input formats while operating within a specialized chat template system. This development is significant for the AI/ML community as it democratizes access to advanced translation tools, enhancing innovation in natural language processing and computer vision applications. The models were fine-tuned on a diverse dataset using advanced Tensor Processing Units (TPUs), which facilitate efficient training and scalability. Additionally, TranslateGemma has undergone rigorous ethical evaluations to reduce biases and improve safety, addressing concerns about content policy adherence across various contexts. By providing both text-to-text and image-to-text translation capabilities, TranslateGemma represents a substantial step forward in the convergence of language and vision technologies.